Path toward testing with a released version of hypercorn? about urllib3 HOT 3 OPEN

Hey @musicinmybrain, thanks for this. And as a Fedora user, thanks for your packaging work! We switched to Hypercorn because it supports HTTP/2 (unlike Tornado), so switching back isn't an option. Sorry that this causes you more work. :( We definitely want to merge our changes upstream, but had little success so far. The urllib3-changes branch contains three important changes for us:

1. Support for the ASGI TLS extension. The current support is partial because it's what urllib3 needed, but I'm happy to contribute complete support. It's computationally expensive, so probably needs to be behind a a configuration option.
2. Access to the underlying transport, which I also implemented as an ASGI extension, though no spec currently exists for this. It was the only way to implement CONNECT, which we need to test our proxy support. It seems reasonable to include to me, and the leading underscore makes it clear that no guarantee exists.
3. Fail loudly in case of h11/h2 errors. The default behavior is to drop errors silently, which makes debugging extremely difficult. I'm happy to put this behind an option too.

@pgjones What do you think about the above? How can we best help you add those features? We can direct some of our HTTP/2 funding to Hypercorn work on those three issues.

from urllib3.

Thank you for writing a full and clear explanation in one place!

I have a working proof of concept for introducing python-hypercorn, python-quart, and python-quart-trio packages to Fedora, and then using PYTHONPATH manipulation to interpose a bundled copy of the forked Hypercorn while running the tests in the python-urllib3 package build. Subject to further review and input from other Fedora packagers, I think that will be the path forward for now.

from urllib3.

1. I'm not sure the ASGI TLS extension can be implemented in full - last time I looked a lot of the aspects weren't possible. I think the ASGI extension specification needs to be rewritten to account for this.
2. & 3. I need to think about these, I had in general not wanted to deviate from the ASGI spec with custom features. I'm more open to 3, as I think the ASGI scope should be simple str/bytes only.

My general issue is time, of which I have very little. PRs are welcome though.

from urllib3.